Early Gait Responsiveness as a Predictor of Functional Trajectory in Older Adults Undergoing Postacute Rehabilitation Authors: Neha Sabharwal Published: 2026 ABSTRACT AbstractObjectiveTo examine early gait responsiveness as a predictor of rehabilitation outcomes in older adults undergoing postacute rehabilitation.DesignRetrospective cohort study.SettingThree skilled nursing facilities in Northern California.ParticipantsOne hundred sixty-two adults aged ≥65 years admitted for postacute rehabilitation (2022-2023); 58% female; primary diagnoses: 38% orthopedic, 33% medical, and 28% neurologic.InterventionsNot applicable (observational study).Main Outcome MeasuresDischarge functional status (Functional Independence Measure motor subscale), length of stay, and discharge disposition.ResultsEarly gait responsiveness was operationalized as the rate of change in gait distance across the first 5 physical therapy sessions. Mean early gait responsiveness was 2.1±1.6 m/session (range −0.6 to 7.4 m/session). Greater early gait responsiveness was independently associated with higher discharge functional status (β=2.6 per m/session; 95% confidence interval, [CI] 1.6-3.6), shorter length of stay (β=−2.0 d per m/session; 95% CI, −2.9 to −1.0), and increased odds of discharge home independently versus institutional placement (odds ratio=2.3; 95% CI 1.4-3.6). The addition of gait responsiveness improved model explanatory power beyond baseline function alone (R² 0.28-0.40).ConclusionsEarly gait responsiveness is a clinically interpretable indicator of rehabilitation trajectory and may support earlier identification of patients at risk for suboptimal outcomes.Keywords: Gait, Prognosis, Rehabilitation, Skilled nursing facility, Subacute care Introduction Despite standardized rehabilitation protocols, older adults demonstrate substantial variability in functional recovery after hospitalization.1, 2, 3 Baseline functional status and therapy dose explain only part of this heterogeneity, and early indicators of recovery trajectory remain poorly defined. In other rehabilitation populations, early response to treatment has not been systematically evaluated as a predictor of longer-term outcomes.4,5 This study focuses on early gait responsiveness, a continuous and clinically interpretable indicator,6, 7, 8 as a potential early signal of functional trajectory. Methods Design and setting This retrospective observational study analyzed deidentified electronic health record data collected between January 2022 and December 2023 from 3 skilled nursing facilities within a single regional rehabilitation network in Northern California. All facilities provided postacute rehabilitation services after acute hospitalization, with physical therapy delivered 5-6 days per week. The study was deemed exempt by the institutional review board as a secondary analysis of deidentified clinical data. Participants Of 214 consecutive admissions screened during the study period, 162 patients aged 65 years or older with at least 5 documented physical therapy sessions were included in the analysis. The cohort was 58% female; primary diagnoses were orthopedic (38%), medical (33%), and neurologic (28%). Exclusion criteria included progressive neurologic disease (eg, amyotrophic lateral sclerosis, Parkinson’s disease; n=15), hospice enrollment at admission (n=8), or incomplete gait distance documentation during the first 5 therapy sessions (n=29). Patients who never achieved ambulation during the initial 5 sessions were also excluded as slope estimation was not feasible. Note that although inclusion required at least 5 documented sessions to standardize the observation window, slope estimation required a minimum of 3 gait measurements; these thresholds serve different analytical purposes. In this study, sex refers to biological sex as documented in the electronic health record; gender identity was not assessed. Gait distance measurement Gait distance was measured during routine physical therapy sessions using facility-standard measurement tools (premarked hallways, measuring wheels, or tape measures) and recorded in meters. Therapists (n=18) across 3 facilities used standardized electronic documentation fields with identical distance recording requirements. Formal inter-rater reliability testing was not conducted, reflecting pragmatic clinical practice in postacute care settings. Patients unable to ambulate at baseline were assigned a gait distance of zero meters for that session. Therapy sessions and missing data Physical therapy sessions were defined as documented visits lasting at least 30 minutes with mobility-related interventions. If gait distance was missing for 1 of the first 5 sessions, slope estimation used the remaining available data points provided that at least 3 gait measurements were present. Patients with fewer than 3 gait measurements were excluded from analysis (n=29). Review of available documentation for excluded patients suggests that incomplete gait recording reflected 2 predominant patterns: (1) early discharge or transfer before 5 sessions were completed (excluded patients had shorter mean lengths of stay than included patients: 11.7 vs 18.7 d, P=.001) and (2) sessions during which patients were medically unavailable (nursing interventions, diagnostic procedures, or patient refusal). No significant difference in age or sex was observed between excluded and included patients. The shorter stays observed in excluded patients suggest data were not missing completely at random; missingness was associated with length of stay, which is itself an outcome variable, potentially biasing the analytical sample toward patients with longer and more complex rehabilitation courses. Slope calculation Early gait responsiveness was operationalized as the rate of change in gait distance (meters) across the first 5 physical therapy sessions. For each patient, ordinary least squares (OLS) linear regression was performed with gait distance (meters) as the dependent variable and session number (1 through 5) as the independent variable. The regression slope coefficient, expressed as meters per session, represented that patient’s early gait responsiveness. The 5-session window was selected a priori to allow timely clinical decision-making (approximately 7-10 d) while providing stable slope estimation with adequate degrees of freedom. Outcomes The primary outcome was discharge functional status, measured using the Functional Independence Measure (FIM) motor subscale as documented at discharge.9,10 The FIM motor subscale ranges from 13 (complete dependence) to 91 (complete independence) across 13 motor items. A change of 22 points is considered the minimal clinically important difference in postacute rehabilitation populations.11 Secondary outcomes included length of stay (d) and discharge disposition categorized as (1) home independent, (2) home with services, or (3) institutional. Statistical analysis Descriptive statistics characterized the sample and key variables. Early gait responsiveness was examined for normality using visual and quantitative assessment. Pearson correlation coefficients assessed the relationship between baseline gait distance and early gait responsiveness. Multivariable linear regression models examined associations between early gait responsiveness and continuous outcomes (discharge FIM motor score, length of stay). Multinomial logistic regression examined associations with discharge disposition, with institutional placement as the reference category. All models adjusted for age, sex, diagnosis category, baseline gait distance, and therapy frequency. To assess the incremental value of early gait responsiveness, 3 nested models were compared for predicting discharge functional status: model 1 (demographic covariates only), model 2 (added baseline gait distance), and model 3 (added early gait responsiveness). Nested linear regression models were compared using F tests; nested multinomial logistic regression models were compared using likelihood ratio tests with chi-squared approximation. Model fit was assessed using R², adjusted R², and Akaike information criterion. Stratified analyses examined whether associations varied across baseline gait tertiles. To assess potential clustering effects from the 3-facility design, intraclass correlation coefficients (ICC) were calculated. Given the low ICC for facility clustering (ICC=0.04), Huber-White heteroscedasticity-consistent sandwich estimators were employed in all regression models to provide conservative inference.12,13 This estimator corrects for heteroscedasticity and mild model misspecification at the patient level without imposing a multilevel generalized estimating equations (GEE) working correlation structure, which was not warranted given the negligible clustering. Regression diagnostics included assessment of residual normality, homoscedasticity, influential observations, and multicollinearity. Statistical significance was set at α=.05 (2-tailed). All analyses were conducted using IBM SPSS Statistics version 29.0.a Although 3 outcomes were examined, alpha levels were not adjusted for multiple comparisons given that outcomes were conceptually correlated components of rehabilitation success. This was an exploratory study, and conservative 2-tailed tests with full reporting of confidence intervals (CIs) provide protection against type I error. Sensitivity analyses Multiple sensitivity analyses assessed robustness of findings: (1) excluding patients with extreme lengths of stay (<10 or >30 d), (2) excluding patients with extreme responsiveness values (>3 SD from mean), (3) diagnosis-specific subgroup analyses, (4) testing for nonlinear relationships; (5) examining alternative responsiveness windows (3 sessions and 7 sessions), and (6) replacing OLS-derived individual slopes with best linear unbiased predictor slopes from a linear mixed effects (LME) model with random intercept and random slope for session, fitted using REML. The LME approach provides more efficient slope estimates by borrowing information across participants, particularly for those with fewer than 5 data points. Sample size The sample size was determined by available data during the study period. Post hoc power analysis confirmed adequate statistical power (>80%) for detecting moderate effects (Cohen’s f² ≥ 0.15) in multivariable linear regression models at α=.05 with n=162.14,15 Missing data Among the 162 included patients, 90.7% (n=147) had gait distance documented at all 5 sessions, 7.4% (n=12) had 4 sessions documented, and 1.9% (n=3) had 3 sessions documented. Missing data were not imputed; slopes were calculated using available data points (minimum 3 required). Sensitivity analyses comparing patients with complete 5-session data to those with 3-4 sessions showed no significant differences in baseline characteristics, suggesting missing data were likely missing at random among the included sample. Results Sample characteristics Participants (n=162) had a mean age of 78.4±7.9 years (range 65-96), 58% female, with primary diagnoses of orthopedic (38%), medical (33%), and neurologic (28%) conditions. Mean baseline gait distance was 18.9±14.6 m (range 0-61.0 m). At discharge, the mean FIM motor score was 64.2±18.4 (range 28-87). Mean length of stay was 18.7±7.4 days (table 1). Table 1Participant characteristics and baseline functional measures (N=162)Table 1 dummy alt textCharacteristicMean±SD or n (%)RangeNotesAge, y78.4±7.965-96Female sex94 (58%)—Primary diagnosis——Orthopedic 38%; medical 33%; neurologic 28%Baseline gait distance, m18.9±14.60-61.0Assistive device allowedBaseline transfer assistance (1-5)2.9±1.11-51=dependent, 5=independentBaseline sit-to-stand reps4.2±2.60-12Therapy sessions/wk4.6±0.83-6Length of stay, d18.7±7.45-45ObservedNote: Transfer assistance coded from 1 (dependent) to 5 (independent). Gait distance recorded in meters (m). Values are presented as mean±standard deviation or number (%). Gait distance is reported in meters. Transfer assistance is coded from 1 (dependent) to 5 (independent). Participant characteristics and baseline functional measures (N=162) Note: Transfer assistance coded from 1 (dependent) to 5 (independent). Gait distance recorded in meters (m). Values are presented as mean±standard deviation or number (%). Gait distance is reported in meters. Transfer assistance is coded from 1 (dependent) to 5 (independent). Patients excluded due to incomplete gait documentation (n=29) did not differ significantly in age or sex but had shorter mean lengths of stay (11.7 vs 18.7 d, P=.001), suggesting potential selection bias toward patients with longer rehabilitation stays. Early gait responsiveness distribution Mean early gait responsiveness was 2.1±1.6 m/session (range −0.6 to 7.4 m/session). Distribution was approximately normal (skewness=0.41, kurtosis=0.22; fig 1A). Twelve patients (7%) demonstrated negative gait responsiveness. Baseline gait distance was moderately negatively correlated with early gait responsiveness (r=−0.32, P<.001), indicating partial regression to the mean. However, substantial variability in responsiveness existed within baseline gait tertiles, suggesting responsiveness captures information beyond initial severity alone. Distribution and association of early gait responsiveness. (A) Distribution of early gait responsiveness (meters per session), demonstrating an approximately normal distribution. (B) Association between early gait responsiveness and discharge functional status (Functional Independence Measure [FIM] motor score), adjusted for age, sex, diagnosis category, baseline gait distance, and therapy frequency. Association with discharge functional status In unadjusted analysis, each 1 m/session increase in early gait responsiveness was associated with a 2.9-point higher discharge FIM motor score (95% CI, 1.9-3.9; P<.001). After adjusting for age, sex, diagnosis, baseline gait distance, and therapy frequency, the association remained significant (β=2.6; 95% CI, 1.6-3.6; P<.001; fig 1B, table 2). Patients in the highest responsiveness quartile (≥3.1 m/session) would be expected to gain approximately 8 points on the FIM motor scale compared to those in the lowest quartile, contributing substantially toward the 22-point minimal clinically important difference.11 Table 2Associations between early gait responsiveness and rehabilitation outcomesTable 2 dummy alt textOutcomeUnadjusted Effect (β or OR) (95% CI)Adjusted Effect (β or OR) (95% CI)P ValueModel FitDischarge FIM motor2.9 (1.9-3.9)2.6 (1.6-3.6)<.001R²=0.40Length of stay, d−2.2 (−3.0 to −1.2)−2.0 (−2.9 to −1.0)<.001R²=0.31Home independent vs institutional2.5 (1.6-3.9)2.3 (1.4-3.6).001Pseudo-R²=0.18Home w/services vs institutional1.9 (1.2-3.0)1.7 (1.1-2.8).02Pseudo-R²=0.18Note: β coefficients expressed per m/session (meters per session) of early gait responsiveness. Adjusted models include age, sex, diagnosis, baseline gait distance, and therapy frequency. LRT for multinomial model: χ²(2)=14.3, P<.001. β coefficients represent the change in outcome per 1 m/session increase in early gait responsiveness. Odds ratios (ORs) are relative to institutional discharge. Adjusted models include age, sex, diagnosis category, baseline gait distance, and therapy frequency.Abbreviations: FIM, Functional Independence Measure; LRT, likelihood ratio test. Associations between early gait responsiveness and rehabilitation outcomes Note: β coefficients expressed per m/session (meters per session) of early gait responsiveness. Adjusted models include age, sex, diagnosis, baseline gait distance, and therapy frequency. LRT for multinomial model: χ²(2)=14.3, P<.001. β coefficients represent the change in outcome per 1 m/session increase in early gait responsiveness. Odds ratios (ORs) are relative to institutional discharge. Adjusted models include age, sex, diagnosis category, baseline gait distance, and therapy frequency. Abbreviations: FIM, Functional Independence Measure; LRT, likelihood ratio test. Nested model comparison Model 1 (demographic covariates only) explained 15% of the variance in discharge FIM scores (R²=0.15). Adding baseline gait distance (model 2) increased explained variance to 28% (R²=0.28, ΔR²=0.13, F test P<.001). Adding early gait responsiveness (model 3) further increased explained variance to 40% (R²=0.40, ΔR²=0.12, F test P<.001). The Akaike information criterion decreased by 34 points from models 2 to 3. Association with length of stay In adjusted models, greater early gait responsiveness was associated with shorter length of stay (β=−2.0 d per m/session; 95% CI, −2.9 to −1.0; P<.001). Patients in the highest responsiveness quartile (≥3.1 m/session) had a mean length of stay of 15.3 days compared with 22.8 days for those in the lowest quartile (<0.9 m/session), a 7.5-day difference (P<.001). Association with discharge disposition In adjusted multinomial logistic regression models, patients in the highest responsiveness quartile had 2.3-fold greater odds of discharge home independently versus institutional placement (odds ratio 2.3; 95% CI, 1.4-3.6; P=.001) and 1.7-fold greater odds of discharge home with services versus institutional placement (odds ratio, 1.7; 95% CI, 1.1-2.8; P=.02; fig 2). The likelihood ratio test for addition of gait responsiveness to the multinomial model was statistically significant (χ²(2)=14.3, P<.001). Association between early gait responsiveness and discharge disposition. Adjusted multinomial logistic regression showing odds of discharge home (independent or with services) relative to institutional placement across levels of early gait responsiveness. Models adjusted for age, sex, diagnosis category, baseline gait distance, and therapy frequency. Stratified analyses Associations between early gait responsiveness and discharge FIM scores remained consistent across baseline gait tertiles: low baseline tertile (β=2.1, P=.003), moderate tertile (β=2.7, P<.001), and high baseline tertile (β=2.4, P=.002). Interaction testing indicated these differences were not statistically significant (interaction P=.42). Patients with negative responsiveness The 12 patients (7%) with negative early gait responsiveness had significantly longer lengths of stay (mean 24.3 vs 17.8 d, P=.03) and lower rates of home discharge (25% vs 68%, P=.001). Facility-level clustering The ICC for facility clustering was 0.04, indicating a minimal clustering effect. Huber-White heteroscedasticity-consistent 1 sandwich estimators were employed; GEE was not warranted given the low ICC. Sensitivity analyses Multiple sensitivity analyses confirmed robustness. Excluding patients with extreme length of stay or responsiveness values did not alter results. Diagnosis-specific effect sizes were similar (orthopedic β=2.7, medical β=2.1, neurologic β=2.4 m/session equivalent; interaction P=.42). Quadratic terms were not significant. Alternative responsiveness windows (3 and 7 sessions) produced consistent findings. The LME sensitivity analysis using best linear unbiased predictor slopes yielded substantively identical associations with discharge FIM motor score (β=2.5 m/session; 95% CI, 1.6-3.4; P<.001), length of stay (β=−1.9 d/m/session, P<.001), and discharge disposition, supporting the robustness of the OLS-based primary analysis. Discussion This study demonstrates that early gait responsiveness was independently associated with functional trajectory in postacute geriatric rehabilitation.1,2,3,9,10,16 The continuous measure avoids psychometric limitations of composite scores while providing clinically interpretable information. Although responsiveness showed moderate negative correlation with baseline gait distance,6, 7, 8 consistent associations across baseline strata suggest it captures meaningful variation beyond initial severity. Mechanistic considerations Early responsiveness likely reflects multiple interconnected factors including cardiopulmonary reserve, neuromuscular plasticity, metabolic capacity, motor learning capacity, and psychosocial factors such as motivation and fear of falling.17 Low responsiveness could also signal unrecognized medical issues requiring intervention.18 The multifactorial nature suggests responsiveness functions as a comprehensive indicator integrating physiological, neurological, medical, psychological, and environmental influences on recovery capacity. Comparison to existing approaches Traditional prognostic tools rely on baseline functional status, age, and comorbidity indices.16, 17, 18, 19, 20, 21, 22 Early gait responsiveness differs by assessing trajectory rather than status, analogous to treatment response paradigms in oncology and psychiatry. Our finding that responsiveness explained an additional 12% of variance beyond baseline function suggests it captures complementary prognostic information. Study limitations This study has important limitations. First, nonstandardized gait distance measurements from routine clinical documentation rather than validated instruments likely introduced measurement variability. Second, approximately 24% of screened patients were excluded due to incomplete documentation or clinical factors, potentially limiting generalizability. Review of excluded patients suggests missingness was associated with shorter length of stay, potentially biasing the included sample toward patients with more complex rehabilitation courses (see Methods section). Third, unmeasured confounding persists, particularly for cognitive status, comorbidity burden, pain, mood, and nutritional status. Fourth, data from 3 facilities within a single regional network in Northern California limit generalizability. External validation in independent cohorts from diverse geographic regions, facility types, payer mixes, and patient populations is essential before widespread clinical adoption. Fifth, measurement by multiple therapists without standardized protocols introduced heterogeneity, although this demonstrates real-world feasibility. Sixth, long-term outcomes beyond discharge were unavailable. Seventh, the 5-session window, although selected a priori, remains somewhat arbitrary. Eighth, participants were enrolled as a consecutive convenience sample of admissions at 3 facilities within a single regional network rather than a probability sample from the target population of older adults undergoing postacute rehabilitation. Consecutive enrollment reduces selection bias within the network but does not ensure representativeness of the broader population. Patients at these facilities may differ systematically from those at other institution types (eg, hospital-based units, urban safety-net facilities, rural facilities) in ways affecting both gait responsiveness distributions and outcome rates. Findings should be interpreted as hypothesis-generating and internally consistent within this sample rather than as generalizable population-level estimates; replication in independently drawn samples from diverse settings is needed before clinical adoption. Finally, this retrospective observational design cannot establish causality. Associations reported here should not be interpreted as causal; early responsiveness may reflect underlying recovery potential rather than treatment effect alone. Study strengths Despite limitations, the sample size provided adequate power for multivariable analyses. Inclusion of 3 facilities enhances generalizability compared with single-center studies. Comprehensive covariate adjustment reduces confounding. The pragmatic design using routine clinical data enhances real-world applicability. Multiple sensitivity analyses confirmed robustness. Clinical implementation The gait responsiveness measure can be readily calculated by rehabilitation staff without specialized equipment. A simplified approximation: responsiveness ≈ (session 5 distance [m] − session 1 distance [m]) ÷ 4. Preliminary cut-points based on this sample’s distribution: low responsiveness (<0.9 m/session, lowest quartile) warranting enhanced monitoring and interdisciplinary review, moderate responsiveness (0.9-3.1 m/session), and high responsiveness (≥3.1 m/session) indicating a favorable prognosis. These thresholds require validation before widespread adoption. Future research Prospective validation in diverse geographic regions and facility types is needed to establish generalizable cut-points. Standardized gait assessments should replace informal distance measurements. Randomized controlled trials comparing responsiveness-guided care to usual care are needed to establish clinical utility and cost-effectiveness. Conclusion Early gait responsiveness provides a simple, actionable indicator of rehabilitation trajectory in older adults. These findings suggest that early within-stay trajectory signals may be clinically actionable before traditional discharge outcomes are observable. Prospective studies are needed to validate thresholds and determine whether responsiveness-guided interventions improve outcomes. Supplier a.IBM SPSS Statistics version 29.0; IBM Corp, Armonk, NY. IBM SPSS Statistics version 29.0; IBM Corp, Armonk, NY. Disclosure The author declares no conflicts of interest. Acknowledgments Author contributions The author was responsible for study conception and design, data analysis and interpretation, manuscript drafting, and final approval of the submitted version. Data statements Due to institutional privacy policies, the de-identified dataset cannot be made publicly available. Requests for aggregate data summaries may be directed to the corresponding author, subject to institutional review board approval, data use agreements, demonstration of legitimate research purpose, and compliance with applicable privacy regulations. Statistical analysis code is available upon request. Ethics approval This study was reviewed by the organization’s institutional review board and determined to be exempt as a secondary analysis of deidentified clinical data. Footnotes List of abbreviations: CI, confidence interval; FIM, Functional Independence Measure; GEE, generalized estimating equations; ICC, intraclass correlation coefficient; LME, linear mixed effects; OLS, ordinary least squares. Supplementary material associated with this article can be found, in the online version, at doi:10.1016/j.arrct.2026.100634. Supplementary materials References Associated Data Supplementary Materials